TL;DR
Thinking Machines, Mistral AI and Microsoft are competing to help enterprises customize models they can treat as their own. Tinker favors portable open weights, Forge emphasizes managed development and European control, while Microsoft ties tuned models closely to Azure.
Thinking Machines, Mistral AI and Microsoft are now selling enterprises three distinct routes to a customized model they can control, following the release of Thinking Machines’ Inkling open weights. The competing offers differ on weight portability, jurisdiction and platform integration, choices that can shape compliance and procurement decisions in healthcare, finance and defense.
Thinking Machines’ Tinker is a low-level training service that lets customers fine-tune open models while the company operates the computing infrastructure. Its supported bases include Inkling, Qwen, DeepSeek, Kimi and other open-weight models. Tinker uses LoRA adapters, and Thinking Machines says customers can download the resulting weights and deploy them elsewhere.
Mistral Forge takes a managed, full-lifecycle approach spanning pre-training and post-training methods such as supervised fine-tuning and reinforcement learning. Mistral presents Forge as a route to customer-owned models that can run on premises, in European infrastructure or in air-gapped environments. That service requires a deeper relationship with Mistral than Tinker’s training interface.
Microsoft’s MAI models and Frontier Tuning, delivered through Foundry, offer weight-level customization inside a large Azure model catalog. Microsoft describes the tuned model as belonging to the customer, but the service carries strong Azure platform dependence. Microsoft has also reported roughly tenfold efficiency gains and cited Mayo Clinic work, though those performance claims have not been independently replicated in the supplied material.
Three ways to own your model: Tinker vs Forge vs Frontier Tuning
Inkling’s open weights were the headline; Tinker is the business. Three serious players now sell the same promise to the same buyer — a model that’s yours, not a rented API — in three different ways. For health, finance & defense, the differences are the whole decision.
For the regulated, defense or health buyer it reduces to one question: what do you most need to control — the weights, the jurisdiction, or the integration? None is strictly best; they’re bets on what you value. The meta-signal: three of the most sophisticated players independently concluded the future enterprise product isn’t a model you rent — it’s one you own and adapt, with your institutional knowledge as the moat. Tinker = portability & open base · Forge = depth & EU sovereignty · Microsoft = lineage & integration. The only wrong move left is renting a generic model and hoping.
Ownership Choices Split Enterprise AI
The contest matters most for organizations whose data cannot move freely because of health-privacy, European data-protection or classification rules. These buyers also need models that can reason within specialized systems such as medical codes, banking regulations or defense signals, rather than merely retrieve domain documents.
The services distribute control differently. Tinker offers the greatest portability and technical independence, but expects customers to supply substantial machine-learning expertise. Forge provides more development support and a European sovereignty case, at the cost of a stickier vendor relationship. Microsoft offers established enterprise integration and model lineage, while making departure from the Azure ecosystem harder.
open weight AI model training platform
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Three Platforms, Three Control Models
The comparison follows Thinking Machines’ release of Inkling, which drew attention because its weights are open. The accompanying commercial logic is Tinker: organizations can start with Inkling or another supported base, customize it through the company’s infrastructure and then export the trained checkpoint.
Mistral and Microsoft are pursuing the same regulated buyers from different positions. Mistral’s offer centers on managed model creation and deployment under European operational control. Microsoft combines first-party MAI models, Frontier Tuning and Foundry’s reported catalog of about 11,000 models. Together, the products indicate growing vendor interest in selling adaptation and governance rather than access to a single general-purpose API.
“Customer data is used only to train the customer’s models, not Thinking Machines’ models.”
— Thinking Machines, according to its Tinker documentation
Vendor Claims Still Need Testing
Several points remain unresolved. The supplied material does not provide independent performance comparisons, audited privacy findings or standardized cost data across the three services. Claims about LoRA matching full fine-tuning, Microsoft’s reported efficiency improvement and each platform’s security posture are vendor-reported.
The practical meaning of model ownership also varies. Downloadable Tinker checkpoints offer a clearer exit path, while Forge customers may depend on Mistral’s development program and Microsoft customers may rely on Azure services. Contract terms, licenses, export rights and support obligations will determine how much control a buyer has in practice.
Procurement Tests Come Next
Enterprise buyers will next need to compare the platforms through domain-specific trials, security reviews and contract analysis. Key tests include whether a model can leave the vendor’s infrastructure, which party controls its weights, how training data is handled and whether deployments can operate on premises or offline.
Independent benchmarks and named production deployments would clarify the vendors’ performance claims. Until those arrive, the choice rests less on a single model score than on whether an organization prioritizes portable weights, European operational control or Azure integration.
Key Questions
Which platform offers the most portable customized model?
Tinker appears to offer the clearest portability because customers can fine-tune supported open models and download the resulting weights. Actual portability still depends on the base model’s license and the customer’s deployment capacity.
How is Mistral Forge different from Tinker?
Forge is a managed development program covering pre-training and post-training, while Tinker exposes lower-level training functions to technical teams. Forge emphasizes European and air-gapped deployment; Tinker emphasizes researcher control and exportable checkpoints.
Do Microsoft customers fully own Frontier-Tuned models?
Microsoft describes the tuned model as the customer’s, but the supplied analysis characterizes deployment as closely tied to Azure. Buyers would need to examine their contracts and technical export options before treating that ownership as equivalent to a portable checkpoint.
Why are regulated industries the main target?
Healthcare, finance and defense organizations face strict data-location and security rules. They also need domain-specific reasoning and clear answers about training-data use, model lineage and deprecation risk.
Has one approach proved better than the others?
No independent comparison supplied here establishes a winner. The three offers optimize for different priorities, and major performance, privacy and cost claims still require outside testing.
Source: Thorsten Meyer AI
